State Of Health Estimation For Lithium-ion Batteries Based On Extraction Of Multiple Feature Variables

The performance of lithium-ion batteries for electric vehicles will deteriorate seriously under the working conditions of high temperature,low temperature and fast charging working conditions.Accurate prediction state of health(SOH)for batteries contributes a lot to battery energy storage system,which can avoid hazards and prolong battery life.In this paper,the lithium-ion batteries cycle life tests are conducted,the data during the charging and discharging phase is collected and the charging and discharging curve and the incremental capacity(IC)curve are drawn.Based on the charging and discharging curve and the IC curve,the characteristic parameters reflecting the aging law of the battery and the health features are acquired and a datadriven model is established to estimate the SOH.Firstly,the parameters and influencing factors of the battery are analyzed from the perspective of the structure and working principle of the lithium-ion battery,and the characteristics of the lithium-ion battery are analyzed;the time vector and the variation of IC during the charging and discharging processes are acquired through cycle life test,and the charge-discharge curve and IC curve are drawn.Secondly,the characteristic parameters reflecting the aging law of batteries are acquired based on the charge-discharge curve and the IC curve.The health features related to battery decay are acquired according to the data of battery aging experiment.The grey relational analysis(GRA)and principal component analysis(PCA)are employed to analyze the characteristic parameters of batteries aging and tackle the problem of redundant or insufficient health factors.Then,the particle swarm optimization(PSO)algorithm is hired to optimize the radial basis function(RBF)neural network(NN)and search the optimization model parameters.Experimental comparisons are carried out on single Elman neural network(Elman NN)and traditional RBF NN model,the results show that the proposed method in this paper can effectively predict the SOH,the maximum error can be limited to 2%,and it features good stability and robustness.Finally,the phenomenon of small capacity regeneration caused by the rest during the cycle life tests,which has a great impact on the local change trend of the SOH is discussed,and a fusion algorithm integrating PSO-RBF NN and long short-term memory neural network(LSTM NN)is proposed to forecast SOH.The original battery health state sequence is processed into three IMFs and a residue through empirical mode decomposition(EMD).The original information is decomposed into high and low frequency parts,the high frequency part includes three intrinsic mode functions(IMFs),and the low frequency part is the residue.The PSO-RBF NN is used to predict the high frequency part,and Adam is used to optimize the parameters of LSTM NN to predict the low frequency part to complete the accurate prediction of the SOH.According to the analysis of the prediction results for different batteries,the maximum error of the SOH can be kept within ±1.5%,thus the proposed fusion algorithm can not only consider the impact of complex changes on the health status,but also ensure high prediction accuracy.